Field of the Invention
The present invention relates to a technique for detecting a visually conspicuous area in a captured image.
Description of the Related Art
In recent years, a technique for detecting, among subjects that are present in an image, a subject that is likely to be conspicuous to human vision has been proposed. For example, in Japanese Patent Laid-Open No. 2012-123631, a visually salient area is detected based on a difference in feature amount between a training data extraction area with a small radius and a verification data extraction area with a large radius. That is, it is determined that the larger the difference in feature amount between the training data extraction area and the verification data extraction area surrounding the training data extraction area is, the higher the degree of visual saliency (hereinafter, referred to as the degree of visual saliency) is.
In the above-described Japanese Patent Laid-Open No. 2012-123631, the sizes of the training data extraction area and the verification data extraction area are determined based on learning data of a subject, and the degree of visual saliency at an arbitrary point in the screen is calculated. However, there is a problem in that the sizes of the training data extraction area and the verification data extraction area cannot be determined in the case of a subject whose learning data does not exist, and the accuracy of the degree of visual saliency deteriorates. There is also a problem in that, for example, if a plurality of areas with different sizes are set and degrees of visual saliency at arbitrary points are calculated, the processing amount will be huge.